ChatGPT Races in the Great Pumpkin Regatta
From helping teenagers write an extraordinary term paper to the elimination of the human race, artificial intelligence has been the hyperbolic cultural and economic phenomenon of the 2020s. Radiologists, Uber drivers and contact center agents all seem to be on the near-term AI chopping block, but what about the sailboat tactician?
Richmond Yacht Club’s Great Pumpkin race is one of the best-attended regattas on the West Coast, with 158 boats registered for the Sunday pursuit race. The format requires rounding both Alcatraz and Angel Island in a single loop, but the entrant can choose the direction. The decision requires a knowledge of the tides, currents, wind conditions and local geography, not to mention basic sailing characteristics. Despite the large number of experienced local sailors, the fleet typically divides into two major groups reflecting a lack of simple consensus.
The author posed this strategic question to OpenAI’s ChatGPT to understand its sophistication on a complex and practical problem. We then had the top two finishing boats review the script and gauge the quality and accuracy of the recommendation.
Ian Charles is the skipper and owner of Maverick, a J/105 competing for St. Francis Yacht Club and finishing in first place. Ian grew up sailing and raced one-design boats ranging from Sunfish and Lasers to multiple J/Boats and Melges. Ian is also a 10-time podium finisher in the Rolex Big Boat Series, and has won multiple one-design regattas in various events.
John Sweeney is the tactician for Will Benedict’s J/105 Advantage 3, representing the Richmond Yacht Club — the second-place finisher. John has competed in three America’s Cup campaigns, serving as main trimmer for America True (2000) and Oracle BMW Racing (2003), and as sporting director for Shosholoza South Africa (2007).
Overall assessment:
Ian: “This was worse than I expected. Beyond it not having access to real-time conditions, it really struggled to put together a coherent course, and as a consequence there is very little usable here. I don’t think AI is capable of answering this question currently; there is not a huge business model for sailing strategy, and it hasn’t received training on this task. As I said earlier, it should have said, ‘I can’t do this,’ rather than hallucinate through the entire exercise. I would give this analysis a 1 out of 10, but interested to see how it improves in the next few years.”
John: “It got the headline correct (it had a 50% chance on this). Pretty much everything else is very confused and/or incorrect. However, it does have a couple pieces of useful information. It just needs to stitch into a coherent narrative. Overall I would rate this analysis 2/10 as an avid AI user. For lawsuits I don’t use chat GPT, except for Google searches. I use Grok for complex problems or law. Zero hallucinations!”
Well, humans win this time, and the tactician’s job appears safe for now, though we all remember when humans used to beat early computers at chess and Go. We’ll check back this time next year.
You can see the prompt and complete commentary from Ian and John about ChatGPT’s efforts as a tactician here.
Who else is using ChatGPT as their tactician? Login to add your comments below.